diff --git a/CHANGELOG.md b/CHANGELOG.md
index d848133fd..3ccef0a38 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -3,7 +3,8 @@
 ## Unreleased — 2025-12-23
 
 ### Fixes
-- Telegram/WhatsApp: native replies now target the original inbound message; reply context is captured in `ReplyTo*` fields for templates. (Thanks @joshp123 for the PR and follow-up question.)
+- Telegram/WhatsApp: native replies now target the original inbound message; reply context is appended to `Body` and captured in `ReplyTo*` fields. (Thanks @joshp123 for the PR and follow-up question.)
+- Embedded agent: custom model providers now load from `models.providers` (merged into `~/.clawdis/agent/models.json`), enabling proxy/base URL setups.
 
 ## 2.0.0-beta2 — 2025-12-21
 
diff --git a/docs/surface.md b/docs/surface.md
index 86107a79a..26827ebc2 100644
--- a/docs/surface.md
+++ b/docs/surface.md
@@ -10,7 +10,7 @@ Updated: 2025-12-07
 Goal: make replies deterministic per channel while keeping one shared context for direct chats.
 
 - **Surfaces** (channel labels): `whatsapp`, `webchat`, `telegram`, `voice`, etc. Add `Surface` to inbound `MsgContext` so templates/agents can log which channel a turn came from. Routing is fixed: replies go back to the origin surface; the model doesn’t choose.
-- **Reply context (optional):** inbound replies may include `ReplyToId`, `ReplyToBody`, and `ReplyToSender` so templates can surface the quoted context when needed.
+- **Reply context:** inbound replies include `ReplyToId`, `ReplyToBody`, and `ReplyToSender`, and the quoted context is appended to `Body` as a `[Replying to ...]` block.
 - **Canonical direct session:** All direct chats collapse into the single `main` session by default (no config needed). Groups stay `group:<jid>`, so they remain isolated.
 - **Session store:** Keys are resolved via `resolveSessionKey(scope, ctx, mainKey)`; the agent JSONL path lives under `~/.clawdis/sessions/<SessionId>.jsonl`.
 - **WebChat:** Always attaches to `main`, loads the full session transcript so desktop reflects cross-surface history, and writes new turns back to the same session.
diff --git a/docs/telegram.md b/docs/telegram.md
index 03c25dad2..3f05b3d28 100644
--- a/docs/telegram.md
+++ b/docs/telegram.md
@@ -34,7 +34,7 @@ Status: ready for bot-mode use with grammY (long-polling by default; webhook sup
 
 ## Planned implementation details
 - Library: grammY is the only client for send + gateway (fetch fallback removed); grammY throttler is enabled by default to stay under Bot API limits.
-- Inbound normalization: maps Bot API updates to `MsgContext` with `Surface: "telegram"`, `ChatType: direct|group`, `SenderName`, `MediaPath`/`MediaType` when attachments arrive, `Timestamp`, and reply-to metadata (`ReplyToId`, `ReplyToBody`, `ReplyToSender`) when the user replies; groups require @bot mention by default.
+- Inbound normalization: maps Bot API updates to `MsgContext` with `Surface: "telegram"`, `ChatType: direct|group`, `SenderName`, `MediaPath`/`MediaType` when attachments arrive, `Timestamp`, and reply-to metadata (`ReplyToId`, `ReplyToBody`, `ReplyToSender`) when the user replies; reply context is appended to `Body` as a `[Replying to ...]` block; groups require @bot mention by default.
 - Outbound: text and media (photo/video/audio/document) with optional caption; chunked to limits. Typing cue sent best-effort.
 - Config: `TELEGRAM_BOT_TOKEN` env or `telegram.botToken` required; `telegram.requireMention`, `telegram.allowFrom`, `telegram.mediaMaxMb`, `telegram.proxy`, `telegram.webhookSecret`, `telegram.webhookUrl`, `telegram.webhookPath` supported.
 
diff --git a/src/telegram/bot.test.ts b/src/telegram/bot.test.ts
index a968329ca..e9f3c2348 100644
--- a/src/telegram/bot.test.ts
+++ b/src/telegram/bot.test.ts
@@ -141,7 +141,8 @@ describe("createTelegramBot", () => {
 
     expect(replySpy).toHaveBeenCalledTimes(1);
     const payload = replySpy.mock.calls[0][0];
-    expect(payload.Body).not.toContain("Reply to Ada: Can you summarize this?");
+    expect(payload.Body).toContain("[Replying to Ada]");
+    expect(payload.Body).toContain("Can you summarize this?");
     expect(payload.ReplyToId).toBe("9001");
     expect(payload.ReplyToBody).toBe("Can you summarize this?");
     expect(payload.ReplyToSender).toBe("Ada");
diff --git a/src/telegram/bot.ts b/src/telegram/bot.ts
index d4dfe7a7b..a83651d8b 100644
--- a/src/telegram/bot.ts
+++ b/src/telegram/bot.ts
@@ -124,13 +124,16 @@ export function createTelegramBot(opts: TelegramBotOptions) {
         ""
       ).trim();
       if (!rawBody) return;
+      const replySuffix = replyTarget
+        ? `\n\n[Replying to ${replyTarget.sender}]\n${replyTarget.body}\n[/Replying]`
+        : "";
       const body = formatAgentEnvelope({
         surface: "Telegram",
         from: isGroup
           ? buildGroupLabel(msg, chatId)
           : buildSenderLabel(msg, chatId),
         timestamp: msg.date ? msg.date * 1000 : undefined,
-        body: rawBody,
+        body: `${rawBody}${replySuffix}`,
       });
 
       const ctxPayload = {
diff --git a/src/web/auto-reply.test.ts b/src/web/auto-reply.test.ts
index 31bb65ce6..640c3fb59 100644
--- a/src/web/auto-reply.test.ts
+++ b/src/web/auto-reply.test.ts
@@ -1786,10 +1786,13 @@ describe("web auto-reply", () => {
       ReplyToId?: string;
       ReplyToBody?: string;
       ReplyToSender?: string;
+      Body?: string;
     };
     expect(callArg.ReplyToId).toBe("q1");
     expect(callArg.ReplyToBody).toBe("original");
     expect(callArg.ReplyToSender).toBe("+1999");
+    expect(callArg.Body).toContain("[Replying to +1999]");
+    expect(callArg.Body).toContain("original");
   });
 
   it("applies responsePrefix to regular replies", async () => {
diff --git a/src/web/auto-reply.ts b/src/web/auto-reply.ts
index 029292a8f..ea17916e8 100644
--- a/src/web/auto-reply.ts
+++ b/src/web/auto-reply.ts
@@ -933,6 +933,12 @@ export async function monitorWebProvider(
 
     const backgroundTasks = new Set<Promise<unknown>>();
 
+    const formatReplyContext = (msg: WebInboundMsg) => {
+      if (!msg.replyToBody) return null;
+      const sender = msg.replyToSender ?? "unknown sender";
+      return `[Replying to ${sender}]\n${msg.replyToBody}\n[/Replying]`;
+    };
+
     const buildLine = (msg: WebInboundMsg) => {
       // Build message prefix: explicit config > default based on allowFrom
       let messagePrefix = cfg.inbound?.messagePrefix;
@@ -945,7 +951,10 @@ export async function monitorWebProvider(
         msg.chatType === "group"
           ? `${msg.senderName ?? msg.senderE164 ?? "Someone"}: `
           : "";
-      const baseLine = `${prefixStr}${senderLabel}${msg.body}`;
+      const replyContext = formatReplyContext(msg);
+      const baseLine = `${prefixStr}${senderLabel}${msg.body}${
+        replyContext ? `\n\n${replyContext}` : ""
+      }`;
 
       // Wrap with standardized envelope for the agent.
       return formatAgentEnvelope({